Exploration
Satyajeet Parida; Abhishek Kumar Tripathi; Tarek Salem Abdennaji; Yewuhalashet Fissha
Abstract
Coal quality is predominantly determined by its Gross Calorific Value (GCV), which directly influences its economic valuation. Traditional empirical formulas for GCV estimation, though effective, become inefficient and laborious when handling large datasets. To address this, machine learning (ML) techniques ...
Read More
Coal quality is predominantly determined by its Gross Calorific Value (GCV), which directly influences its economic valuation. Traditional empirical formulas for GCV estimation, though effective, become inefficient and laborious when handling large datasets. To address this, machine learning (ML) techniques offer a robust alternative for accurate and rapid predictions. This study employs seven coal quality parameters. Total Moisture (TM), Ash (ASH), Volatile Matter (VM), Hydrogen (H), Carbon (C), Nitrogen (N), and Sulphur (S), as independent variables to develop predictive models for GCV. Four conventional regression techniques, namely Support Vector Regression (SVR), K-Nearest Neighbors (KNN), Random Forest (RF), and Decision Tree (DT), along with two robust regression models Random Sample Consensus (RANSAC) and Huber Regressor (HR) are explored. The dataset comprises coal samples from five Asia-Pacific countries: China, Indonesia, Korea, the Philippines, and Thailand. Comparative performance analysis reveals that the robust regression models significantly outperform the conventional ML techniques. The RANSAC and Huber Regressor models achieve superior prediction accuracy with R² values of 0.9941 and 0.9952, respectively. These findings highlight the potential of robust regression approaches for reliable GCV estimation, facilitating efficient coal quality assessment in large-scale applications.
